arXiv subjects

Yihang Du

Publications and source records attributed to Yihang Du.

5 recordsLinked to original sources

arXiv2026

Why Vision Fails as a Universal Bridge: Rectifying Modality Asynchrony in Multilingual MLLMs

Multimodal large language models (MLLMs) exhibit substantial performance degradation in non-English visual reasoning, despite the strong multilingual competence of their text-only backbones. While mechanistic evidence from text-only models suggests that non-English inputs are routed through an English-centric latent space, the multimodal implications of this phenomenon remain unexplored. Through rigorous mechanistic analysis, we identify the \textbf{Ghost Anchor} phenomenon: a temporal modality asynchrony where linguistic translation to the English semantic manifold completes in early layers, while visual semanticization remains immature. Consequently, visual signals are physically present yet functionally invisible during the early alignment window. To rectify this, we propose \textbf{ANCHOR}, a training framework employing Proactive Visual Anchoring (PVA) to accelerate early visual semantic emergence, ensuring visual representations proactively guide linguistic translation. Mechanistic interventions confirm that ANCHOR successfully restores the causal influence of visual signals during early translation. Furthermore, extensive experiments on XMMMU, MaXM, and CVQA demonstrate that ANCHOR consistently outperforms standard baselines, achieving robust visual reasoning across both fine-tuned and zero-shot languages.

cs.CL↗

arXiv2025

Causality-Induced Positional Encoding for Transformer-Based Representation Learning of Non-Sequential Features

Positional encoding is essential for supplementing transformer with positional information of tokens. Existing positional encoding methods demand predefined token/feature order, rendering them unsuitable for real-world data with non-sequential yet causally-related features. To address this limitation, we propose CAPE, a novel method that identifies underlying causal structure over non-sequential features as a weighted directed acyclic graph (DAG) using generalized structural equation modeling. The DAG is then embedded in hyperbolic space where its geometric structure is well-preserved using a hyperboloid model-based approach that effectively captures two important causal graph properties (causal strength & causal specificity). This step yields causality-aware positional encodings for the features, which are converted into their rotary form for integrating with transformer's self-attention mechanism. Theoretical analysis reveals that CAPE-generated rotary positional encodings possess three valuable properties for enhanced self-attention, including causal distance-induced attenuation, causal generality-induced attenuation, and robustness to positional disturbances. We evaluate CAPE over both synthetic and real-word datasets, empirically demonstrating its theoretical properties and effectiveness in enhancing transformer for data with non-sequential features. Our code is available at https://github.com/Catchxu/CAPE.

cs.LG↗

arXiv2025

A Methodological Framework for Measuring Spatial Labeling Similarity

Spatial labeling assigns labels to specific spatial locations to characterize their spatial properties and relationships, with broad applications in scientific research and practice. Measuring the similarity between two spatial labelings is essential for understanding their differences and the contributing factors, such as changes in location properties or labeling methods. An adequate and unbiased measurement of spatial labeling similarity should consider the number of matched labels (label agreement), the topology of spatial label distribution, and the heterogeneous impacts of mismatched labels. However, existing methods often fail to account for all these aspects. To address this gap, we propose a methodological framework to guide the development of methods that meet these requirements. Given two spatial labelings, the framework transforms them into graphs based on location organization, labels, and attributes (e.g., location significance). The distributions of their graph attributes are then extracted, enabling an efficient computation of distributional discrepancy to reflect the dissimilarity level between the two labelings. We further provide a concrete implementation of this framework, termed Spatial Labeling Analogy Metric (SLAM), along with an analysis of its theoretical foundation, for evaluating spatial labeling results in spatial transcriptomics (ST) \textit{as per} their similarity with ground truth labeling. Through a series of carefully designed experimental cases involving both simulated and real ST data, we demonstrate that SLAM provides a comprehensive and accurate reflection of labeling quality compared to other well-established evaluation metrics. Our code is available at https://github.com/YihDu/SLAM.

cs.LG↗

arXiv2022

Underdetermined 2D-DOD and 2D-DOA Estimation for Bistatic Coprime EMVS-MIMO Radar: From the Difference Coarray Perspective

In this paper, the underdetermined 2D-DOD and 2D-DOA estimation for bistatic coprime EMVS-MIMO radar is considered. Firstly, a 5-D tensor model was constructed by using the multi-dimensional space-time characteristics of the received data. Then, an 8-D tensor has been obtained by using the auto-correlation calculation. To obtain the difference coarrays of transmit and receive EMVS, the de-coupling process between the spatial response of EMVS and the steering vector is inevitable. Thus, a new 6-D tensor can be constructed via the tensor permutation and the generalized tensorization of the canonical polyadic decomposition. {According} to the theory of the Tensor-Matrix Product operation, the duplicated elements in the difference coarrays can be removed by the utilization of two designed selection matrices. Due to the centrosymmetric geometry of the difference coarrays, two DFT beamspace matrices were subsequently designed to convert the complex steering matrices into the real-valued ones, whose advantage is to improve the estimation accuracy of the 2D-DODs and 2D-DOAs. Afterwards, a third-order tensor with the third-way fixed at 36 was constructed and the Parallel Factor algorithm was deployed, which can yield the closed-form automatically paired 2D-DOD and 2D-DOA estimation. The simulation results show that the proposed algorithm can exhibit superior estimation performance for the underdetermined 2D-DOD and 2D-DOA estimation.

eess.SP↗

arXiv2022

8D Parameters Estimation for Bistatic EMVS-MIMO Radar via the nested PARAFAC

In this letter, a novel nested PARAFAC algorithm was proposed to improve the 8D parameters estimation performance for the bistatic EMVS-MIMO radar. Firstly, the outer part PARAFAC algorithm was carried out to estimate the receive spatial response matrix and its first way factor matrix. For the estimated first way factor matrix, a theory is given to rearrange its data into an new matrix, which is the mode-1 unfolding matrix of a three-way tensor. Then, the inner part PARAFAC algorithm was used to estimate the transmit steering vector matrix, the transmit spatial response matrix and the receive steering vector matrix. Thus, the transmit 4D parameters and receive 4D parameters can be accurately located via the abovementioned process. Compared with the original PARAFAC algorithm, the proposed nested PARAFAC algorithm can avoid additional reconstruction process when estimating the transmit/receive spatial response matrix. Moreover, the proposed algorithm can offer a highly-accurate 8D parameters estimaiton than that of the original PARAFAC algorithm. Simulated results verify the effectiveness of the proposed algorithm.

eess.SP↗